summaryrefslogtreecommitdiffstats
path: root/src/common/x64/native_clock.cpp
diff options
context:
space:
mode:
Diffstat (limited to 'src/common/x64/native_clock.cpp')
-rw-r--r--src/common/x64/native_clock.cpp61
1 files changed, 41 insertions, 20 deletions
diff --git a/src/common/x64/native_clock.cpp b/src/common/x64/native_clock.cpp
index 91b842829..7fd9d22f8 100644
--- a/src/common/x64/native_clock.cpp
+++ b/src/common/x64/native_clock.cpp
@@ -4,33 +4,55 @@
#include <array>
#include <chrono>
-#include <limits>
-#include <mutex>
#include <thread>
#include "common/atomic_ops.h"
#include "common/uint128.h"
#include "common/x64/native_clock.h"
+#ifdef _MSC_VER
+#include <intrin.h>
+#endif
+
namespace Common {
+#ifdef _MSC_VER
+__forceinline static u64 FencedRDTSC() {
+ _mm_lfence();
+ _ReadWriteBarrier();
+ const u64 result = __rdtsc();
+ _mm_lfence();
+ _ReadWriteBarrier();
+ return result;
+}
+#else
+static u64 FencedRDTSC() {
+ u64 result;
+ asm volatile("lfence\n\t"
+ "rdtsc\n\t"
+ "shl $32, %%rdx\n\t"
+ "or %%rdx, %0\n\t"
+ "lfence"
+ : "=a"(result)
+ :
+ : "rdx", "memory", "cc");
+ return result;
+}
+#endif
+
u64 EstimateRDTSCFrequency() {
// Discard the first result measuring the rdtsc.
- _mm_mfence();
- __rdtsc();
+ FencedRDTSC();
std::this_thread::sleep_for(std::chrono::milliseconds{1});
- _mm_mfence();
- __rdtsc();
+ FencedRDTSC();
// Get the current time.
const auto start_time = std::chrono::steady_clock::now();
- _mm_mfence();
- const u64 tsc_start = __rdtsc();
+ const u64 tsc_start = FencedRDTSC();
// Wait for 200 milliseconds.
std::this_thread::sleep_for(std::chrono::milliseconds{200});
const auto end_time = std::chrono::steady_clock::now();
- _mm_mfence();
- const u64 tsc_end = __rdtsc();
+ const u64 tsc_end = FencedRDTSC();
// Calculate differences.
const u64 timer_diff = static_cast<u64>(
std::chrono::duration_cast<std::chrono::nanoseconds>(end_time - start_time).count());
@@ -44,8 +66,7 @@ NativeClock::NativeClock(u64 emulated_cpu_frequency_, u64 emulated_clock_frequen
u64 rtsc_frequency_)
: WallClock(emulated_cpu_frequency_, emulated_clock_frequency_, true), rtsc_frequency{
rtsc_frequency_} {
- _mm_mfence();
- time_point.inner.last_measure = __rdtsc();
+ time_point.inner.last_measure = FencedRDTSC();
time_point.inner.accumulated_ticks = 0U;
ns_rtsc_factor = GetFixedPoint64Factor(NS_RATIO, rtsc_frequency);
us_rtsc_factor = GetFixedPoint64Factor(US_RATIO, rtsc_frequency);
@@ -57,10 +78,10 @@ NativeClock::NativeClock(u64 emulated_cpu_frequency_, u64 emulated_clock_frequen
u64 NativeClock::GetRTSC() {
TimePoint new_time_point{};
TimePoint current_time_point{};
+
+ current_time_point.pack = Common::AtomicLoad128(time_point.pack.data());
do {
- current_time_point.pack = time_point.pack;
- _mm_mfence();
- const u64 current_measure = __rdtsc();
+ const u64 current_measure = FencedRDTSC();
u64 diff = current_measure - current_time_point.inner.last_measure;
diff = diff & ~static_cast<u64>(static_cast<s64>(diff) >> 63); // max(diff, 0)
new_time_point.inner.last_measure = current_measure > current_time_point.inner.last_measure
@@ -68,7 +89,7 @@ u64 NativeClock::GetRTSC() {
: current_time_point.inner.last_measure;
new_time_point.inner.accumulated_ticks = current_time_point.inner.accumulated_ticks + diff;
} while (!Common::AtomicCompareAndSwap(time_point.pack.data(), new_time_point.pack,
- current_time_point.pack));
+ current_time_point.pack, current_time_point.pack));
/// The clock cannot be more precise than the guest timer, remove the lower bits
return new_time_point.inner.accumulated_ticks & inaccuracy_mask;
}
@@ -77,13 +98,13 @@ void NativeClock::Pause(bool is_paused) {
if (!is_paused) {
TimePoint current_time_point{};
TimePoint new_time_point{};
+
+ current_time_point.pack = Common::AtomicLoad128(time_point.pack.data());
do {
- current_time_point.pack = time_point.pack;
new_time_point.pack = current_time_point.pack;
- _mm_mfence();
- new_time_point.inner.last_measure = __rdtsc();
+ new_time_point.inner.last_measure = FencedRDTSC();
} while (!Common::AtomicCompareAndSwap(time_point.pack.data(), new_time_point.pack,
- current_time_point.pack));
+ current_time_point.pack, current_time_point.pack));
}
}